Schools devote substantial resources to teacher professional development each year. Yet studies show much of this investment is directed toward ineffective short-term workshops that have little impact on instructional change or student outcomes. At the same time, more intensive job-embedded forms of professional learning, such as instructional coaching, require substantially more resources than traditional professional development. We report the results of a two-year study assessing the cost-effectiveness of instructional coaching through a design-based, continuous improvement research model. We study iterative, inquiry cycles in which educators collect data and make changes to the coaching model based on multiple rounds of implementation. We determined the effectiveness of coaching during each iteration by tracking the number of times teachers and coaches reached student-outcome based goals set during the coaching cycle. We assess the cost of implementing the coaching model for each of the three iterations by monitoring staff time allocations and other resource use. Results show that across five schools, the cost of the coaching intervention decreased substantially from the first iteration to the second iteration but increased moderately during the third iteration. Our findings suggest that coaching programs can become more cost-effective over time, as coaches and teachers refine their work together. While specific design features of the study limit generalizability of our findings, the study demonstrates how improvement science or design-based research can be combined with cost-effectiveness research to improve practice in local settings.